1. SAR detection and model tracking of oil slicks in the Gulf of Mexico
Xiaofeng Li
NOAA/NESDIS
Contributors:
William Pichel, NOAA, 5200 Auth Road, Room 102, Camp Springs, MD, 20746, USA
Biao Zhang and Will Perrie, Bedford Institute of Oceanography, Dartmouth, CANADA
Oscar Garcia, Florida State University, 117 N. Woodward Avenue, Tallahassee, FL, 32306, USA
Yongcun Cheng, Danish National Space Center, DTU, DK-2100, Copenhagen, Denmark
Peng Liu, George Mason University

2. Outline Oil Spill Detection in SAR image
Tracking of oil spill movement in the Gulf of Mexico
Deepwater Horizon Event –
NESDIS Effort to Map Surface Oil with Satellite SAR

3. 1. Oil Slicks Detection with SAR
Oil detection with image data and complex data:
1.1 Oil detection with single-pol SAR image
1.2. A Multi-Pol SAR processing chain to observe oil fields
January, 2009

4. 1.1 Oil Slicks Detection with single-polSAR image
Mechanism:
Oil slick damp the ocean surface capillary waves – making the surface smoother
The smooth surface will reflect the radar pulse in the forward direction -> Less backscatter. Radar image is dark.
Challenge:
There are a lot of look-alikes in the SAR image, i.e., low wind, coastal upwelling, island shadow, rain cell, biogenic slicks, etc.
Solution:
Statistical method to extract oil slick from the SAR image
Separate the look-alikes from the oil slick

5. Neural Network Algorithm
1.1 Oil Slicks Detection with single-polSAR image- Algorithms
Neural Network Algorithm
Canadian Journal of Remote Sensing, Vol 25, No

6. 1.1 Oil Slicks Detection with single-polSAR image- Algorithms
8bit pixel value
Wind Magnitud
Wind Direction
Wind Magnitud (-3 h)
Wind Direction (-3 h)
Wind Magnitud (-6 h)
Wind Direction (-6 h)
Wind Magnitud (-9 h)
Wind Direction (-9 h)
Beam Mode Incidence Angle
Sea Surface Height
Geostrophic Currents Magnitud
Geostrophic Currents Direction
Neighboor Texture 1 (Brightness)
Neighboor Texture 2 (Contrast)
Neighboor Texture 3 (Distribution)
Neighboor Texture 4 (Entropy)
Neighboor Texture 5 (variability)
Neighboor Texture 6 (Std Deviation)
1st Filter Reaction
2nd Filter Reaction
3rd Filter Reaction
4th Filter Reaction
5th Filter Reaction
6th Filter Reaction
7th Filter Reaction
8th Filter Reaction
9th Filter Reaction
1.1 Oil Slicks Detection with single-polSAR image- Algorithms
Slick
No-Slick
Neural Network Algorithm demo

7. 1.1 Oil Slicks Detection with single-polSAR image- Results

8. 1.1 Oil Slicks Detection with single-polSAR image- Results

9. 1.1 Oil Slicks Detection with single-polSAR image- Results in GIS

11. In this example, Monitoring BP oil spill
a SAR image was collected by Envisat on June 9, 2010.
Oil is detected close to Louisiana peninsula.
TCNNA now has been trained to process SAR data from:
-RADARSAT 1-2
ENVISAT
ALOS

12. TCNNA GUI: Display of a a pre-processed output.
This Window of the GUI shows wind conditions prevailing
on the data from CMOD5 model.
The TCNNA Output is exported with its
Geo-referenced tagged information.
Ready for Arcmap.
A scaled image is rotated and shown
to adjust contrast along incidence angles

13. TCNNA output handled and converted to Shapefile in ArcMap or Kml for Google Earth

14. 1.1 Single-Pol SAR oil detection summary
Statistical-based SAR oil detection algorithms are developed
These algorithm are tuned for RADARSTA-1, ENVISAT, ALOS, ERS in various beam mode
Interactive oil spill analysis software have been developed to aid oil spill analysis at NOAA

15. The combination of polarimetric features extraction
1.2. A Multi-Polarimetric SAR Processing Chain to Observe Oil Fields in the Gulf of Mexico
The combination of polarimetric features extraction
Total power span image
Co-polar correlation coefficient
Target Decomposition
entropy (H)
mean scattering angle (α)
anisotropy A
The combined feature F

16. PolSAR sea surface scattering
Sea surface (Rough)
Bragg scattering
Low pol.entropy
High HH VV correlation
Oil spill (Smooth)
Non Bragg scattering
High pol. entropy
Low HH VV correlation

17. Example with: NASA UAVSAR polarimetric L-band SAR, with range resolution of 2 m and a range swath greater than 16 km, June 23, :42 (UTC)
A sub scene of UAVSAR image
The image recorded by a video camera
confirmed the oil spill.

18. Extracted polarimetric features from the UAVSAR data

19. The combined polarimetric features and the result of OTSU segmentation

20. Case 2: RADARSAT-2 Oil slick observation
Imaging mode: fine quad-pol SLC
Azimuth pixel spacing: 4.95 m
Range pixel spacing: 4.73 m
Near range incidence: 41.9 degree
Far range incidence: 43.3 degree
Noise floor: ~ -36 dB VV HH
R2 fine quad-pol SAR image of oil slicks in the GOM acquired at 12:01 UTC May 8, 2010

21. Oil slick-covered area
Case 2: RADARSAT-2 Oil slick observation
Clean sea surface
Oil slick-covered area
Under moderate radar incidence angles
and wind speeds
Capillary and small gravity waves were damped
Surface Bragg scattering
Non-Bragg scattering

22. R2 quad-pol observations
Case 2: RADARSAT-2 Oil slick observation
R2 quad-pol observations
scattering matrix
entropy
alpha
represent and characterize scattering mechanism

23. Entropy represents randomness of scattering mechanism
Case 2: RADARSAT-2 Oil slick observation
Entropy represents randomness of scattering mechanism
Entropy low
Entropy high
significant
polarimetric information
backscatter becomes
depolarized
Surface Bragg scattering
Non-Bragg scattering

24. Alpha angle characterizes scattering mechanism
Case 2: RADARSAT-2 Oil slick observation
Alpha angle characterizes scattering mechanism
Surface Bragg scattering dominates
Dipole scattering dominates
Even-bounce scattering dominates
Non-Bragg scattering
Bragg scattering

25. CP for quad-polarization:
Case 2: RADARSAT-2 Oil slick observation
CP for quad-polarization:
For ocean surface Bragg scattering
For non-Bragg scattering
is small
and
have low correlation
and
highly correlated
phase difference is close to
phase difference is close to

26. Case 2: RADARSAT-2 Oil slick observation

27. Case 2: RADARSAT-2 Oil slick observation
Zhang, B., W. Perrie, X. Li, and W. G. Pichel (2011), Mapping sea surface oil slicks using RADARSAT-2 quad-polarization
SAR image, Geophys. Res. Lett., 38, L10602, doi: /2011GL

28. 1.2. A Multi-Polarimetric SAR Processing Chain to Observe Oil Fields in the Gulf of Mexico - Summary
Experimental results demonstrate the physically-based and computer-time efficiency of the two proposed approaches for both oil slicks and man-made metallic targets detection purposes, taking full advantage of full-polarimetric and full-resolution L-band ALOS PALSAR SAR data.
Moreover, the proposed approaches are operationally interesting since they can be blended in a simple and very effective processing chain which is able to both detect and distinguish oil slicks and manmade metallic targets in polarimetric SAR data.

29. 2. Tracking of oil spill movement in the Gulf of Mexico
Introduction to NOAA GNOME Oil drifting model
GNOME Simulation
Simulation results – case study
Conclusions
Main impacts are: - harm to life, property and commerce - environmental degradation

30. 2. Tracking of oil spill movement in the Gulf of Mexico
Oil Slicks drifting simulation with GNOME model
GNOME (General NOAA Operational Modeling Environment) is the oil spill trajectory model used by NOAA’s Office of Response and Restoration (OR&R) Emergency Response Division (ERD) responders during an oil spill. ERD trajectory modelers use GNOME in Diagnostic Mode to set up custom scenarios quickly.
NOAA OR&R employs GNOME as a nowcast/forecast model primarily in pollution transport analyses.
GNOME can:
predict how wind, currents, and other processes might move and spread oil
spilled on the water.
learn how predicted oil trajectories are affected by inexactness ("uncertainty") in current and wind observations and forecasts.
see how spilled oil is predicted to change chemically and physically ("weather") during the time that it remains on the water surface.

31. GNOME input:
- Location file, specific for each region (tide, bathymetry ,etc.)
User file
Currents: ocean model outputs
Winds: model or buoy wind
Oil information: Oil locations from SAR image

32. Spill Trajectory Types
Model Output
Spill Trajectory Types
Best Guess Trajectory (Black Splots)
Spill trajectory that assumes all environmental data and forecasts are correct. This is where we think the oil will go.
Minimum Regret Trajectory (Red Splots) Summary of uncertainty in spill trajectories from possible errors in environmental data and forecasts. This is where else the oil could go.

33. Case study: Oil pipeline leak in July 2009

34. Oil Pipeline leaking in July 2009

39. Oil pipeline leak in July 2009
Surface Currents:
Navy Coastal Ocean Model (NCOM) outputs
spatial resolution of NCOM is 1/8º
temporal resolution is 3 hours

40. Oil pipeline leak in July 2009
Winds:
NDBC hourly wind vector

41. Oil pipeline leak in July 2009
Initial Oil distribution information: denoted by blue dots.
Model run: 7/26/ :00 UTC
7/29/ :00 UTC

42. Simulation Results:
GNOME simulated best guess trajectory of oil spill denoted by blue circles:
16:30 UTC on July 27, 2009
At the ending of the simulation,
04:00 UTC on July 29, 2009.

43. Simulation Results:
GNOME simulated best guess trajectory of oil spill denoted by blue circles:
GNOME simulated locations of the oil spill at 04:00 UTC on July 29, 2009:
only use wind to force the model;
only use the currents to force the model.

44. 2. Tracking of oil spill movement in the Gulf of Mexico - Summary
In this work, the GNOME model was used to simulate an oil spill accident in the Gulf of Mexico. The ocean current fields from NCOM and wind fields measured from NDBC buoy station were used to force the model. The oil spill observations from ENVISAT ASAR and ALOS SAR images were used to determine the initial oil spill information and verify the simulation results. The comparisons at different time show good agreements between model simulation and SAR observations.
Marine Pollution Bulletin, 2010

45. Operational Response Requires:
Summary:
SAR images from multiplatform spaceborne SAR satellite can be used for oil spill/seep detection in the Gulf of Mexico.
Statistical-based oil spill detection algorithms have been developed for single-pol SAR image. These algorithms have been tuned for different satellites and different imaging mode.
A Multi-Frequency Polarimetric SAR Processing Chain to Observe Oil Fields in the Gulf of Mexico are also developed to provide fast oil spill response at NOAA.
The oil spill drifting can be simulated using the NOAA GNOME model with inputs from background current field, time series of wind measurement, and the initial oil spill location.
Operational Response Requires:
SAR is primary data, visible Sun glint secondary, others tertiary
Need multiple looks per day received within 1-2 hours
Many sources of data are required
Well-trained staff of analysts (10-12) to cover multiple shifts per day
Automated mapping would be useful for complicated spill patterns
Array of model, in situ, and complementary imagery and products help by providing an oceanographic context.
Wish for the Future:
What if SAR data were available like this all the time at no per-image cost; i.e., just like most other satellite remote sensing data?